Abstract
Crowding is the disruptive effect of clutter on object recognition. Despite affecting a range of visual features, from colour to motion, most models characterise it as a singular mechanism. Studies using faces have challenged this idea by proposing a separate ‘holistic’ crowding stage (Louie, Bressler & Whitney, 2007). We first replicate the holistic effect using an identity-matching task: crowding is strong for an upright target face (processed ‘holistically’) when surrounded by upright flanker faces and weak when surrounded by inverted flankers. We find no such modulation of crowding by flanker orientation for an inverted target face (processed ‘featurally’). Although this effect has been attributed to tuning for ‘holistic similarity’, we propose that the disproportionate difficulty with inverted faces may obscure lower-level orientation selectivity. In order to match task difficulty for upright and inverted faces, we used an eye-judgment task requiring the detection of horizontal eye-shifts that is immune to inversion effects (Goffaux & Rossion, 2007). Here, we find that crowding is modulated by flanker orientation for both upright and inverted target faces, regardless of their ‘holistic similarity’: crowding is strong when target-flanker orientations match and weak when they differ. The orientation selectivity for inverted faces can nonetheless be ‘switched off’ again with the re-introduction of an inversion effect using a more difficult vertical eye-shift task. Finally, we demonstrate the cause of this modulation of crowding using Thatcherised faces. When the position of flanker facial features is shifted to resemble an inverted face (but without feature rotations), crowding is strongly reduced. In contrast, rotations of the flanker features (leaving upright feature positions intact) have no effect. Our findings demonstrate that the crowding of faces is determined by low-level differences in facial feature positions and not by ‘holistic similarity’, supporting a singular account of crowding.
Meeting abstract presented at VSS 2015